Production monitoring · Autessa Evaluations
Multi-agent evaluation
Autessa Evaluations evaluates multi-agent systems. Manager agents that delegate to sub-agents are judged on the full history and every user request, and Multi-Agent History shows each delegation, the instruction the manager gave and the sub-agent’s own nested transcript.
The problem
In a multi-agent system, who got it wrong?
When an orchestrated answer is bad, the fault could be the manager’s choice of sub-agent, the instruction it wrote, or the sub-agent’s own tool calls. Without the full chain, teams fix the wrong thing.
What you get
- Run rows show evaluation status, close date, manager version, message count, Agents Used and Iterations.
- The orchestration viewer shows the manager’s reasoning before each delegation.
- Each invocation shows the sub-agent, the task it was given, its response summary and conversation ID.
- Expand Agent Conversation Context for the sub-agent’s full nested transcript and tool calls.
- An evaluation side panel with automated scores and human evaluations that take precedence.
- Managers can be tested in the streaming test window, with each sub-agent delegation visible.
How it works
How does multi-agent evaluation work in Autessa Evaluations?
Open Multi-Agent History
Pick the manager and find the run; unsuccessful runs are marked in red.
Check the delegation
Was the right sub-agent chosen, and was the task phrased clearly? If not, fix the manager’s instructions.
Check the sub-agent
Expand its context to read its transcript and tool calls. A failure there is fixed on the sub-agent.
Record a verdict
Add a human evaluation with notes so the run counts correctly.
Every option
What each run shows
| Option | What it does |
|---|---|
| Agents Used | How many sub-agent invocations occurred. |
| Iterations | How many orchestration cycles the manager ran. |
| Invocation cards | Sub-agent, task, response summary, conversation ID and nested context. |
| Evaluation panel | Automated scores plus human evaluations. |
Permissions
Who can do what?
| Who | Can |
|---|---|
| Read | View the agent and its versions, test-run it, and read its history and results. |
| Write | Everything Read can, plus edit the draft and record human evaluations. |
| Publish | Publish and roll back versions, which is what customers get. |
| Admin | Share and delete the agent. The owner is always an Admin. |
FAQ
Questions about multi-agent evaluation
Can Autessa evaluate multi-agent systems?
Yes. Manager conversations are judged on the full history and every user request, with each sub-agent’s transcript available for review.
How do I tell whether the manager or a sub-agent failed?
Read the manager’s reasoning before the delegation first; if it was right, expand the sub-agent’s context and check its own transcript and tool calls.
Do sub-agents show their tool calls?
Yes. The nested Agent Conversation Context includes the sub-agent’s tool requests and responses.
100% customizable, code and all
Make Autessa Evaluations fully your own.
Use it out of the box, then change anything. Ask Autobot, Autessa’s AI builder, for a new field, screen, report, workflow or agent and it builds it into your copy of the app. Or open the code and edit it yourself. Nothing is locked: the app, its data and its code are yours.
Start free and make it yours